The ultimate seizure outcome of childhood epilepsy is complete resolution of all seizures without further treatment. How often this happens and how well it can be predicted early in the course of epilepsy could be valuable in helping families understand the nature of childhood epilepsy and what to expect over time. In the Connecticut study of epilepsy, a prospective cohort of 613 children with newly-diagnosed epilepsy (onset age 0-15 years), complete remission, >/=5 years both seizure-free and medication-free, was examined as a proxy of complete seizure resolution. Predictors at initial diagnosis were tested. Information about seizure outcomes within 2 years and from 2-5 years after diagnosis was sequentially added in a proportional hazards model. The predictive value of the models was determined with logistic regression. Five hundred and sixteen subjects were followed >/=10 years. Three hundred and twenty-eight (63%) achieved complete remission; 23 relapsed. The relapse rate was 8.2 per 1000 person-years and decreased over time: 10.7, 6.7, and 0 during first 5 years, the next 5 years, and then >10 years after complete remission (P = 0.06 for trend). Six participants regained complete remission; 311 (60%) were in complete remission at last contact. Baseline factors predicting against complete remission at last contact included onset age >/=10 years (hazard ratio = 0.55, P = 0.0009) and early school or developmental problems (hazard ratio = 0.74, P = 0.01). Factors predicting for complete remission were uncomplicated epilepsy presentation (hazard ratio = 2.23, P < 0.0001), focal self-limited epilepsy syndrome (hazard ratio = 2.13, P < 0.0001), and uncharacterized epilepsy (hazard ratio = 1.61, P = 0.04). Remission (hazard ratio = 1.95, P < 0.0001) and pharmaco-resistance (hazard ratio = 0.33, P < 0.0001) by 2 years respectfully predicted in favour and against complete remission. From 2 to 5 years after diagnosis, relapse (hazard ratio = 0.21, P < 0.0001) and late pharmaco-resistance (hazard ratio = 0.21, P = 0.008) decreased and late remission (hazard ratio = 2.40, P < 0.0001) increased chances of entering complete remission. The overall accuracy of the models increased from 72% (baseline information only), to 77% and 85% with addition of 2-year and 5-year outcomes. Relapses after complete remission are rare making this an acceptable proxy for complete seizure resolution. Complete remission after nearly 20 years is reasonably well predicted within 5 years of initial diagnosis.